A thumbnail studio that runs on your AI account.
Re-Prompter opens in the Chrome side panel and takes an idea to a finished thumbnail: a creative brief you can read and edit, artwork from your image model, a headline typeset with real letters, and titles to go with it. No credits, no queue, no server of ours in the middle.
// stage 1 of 8 — read it before you pay for an image { "subject": "a trader at a desk, chest up, left third", "lighting": "hard key from camera right, deep falloff", "colourPalette":"near-black, amber, one cold accent", "emotion": "tense, mid-reaction", "textPlacement":"right half kept clean for the headline", "negativePrompt":"text, letters, watermark, logo" }
The part nobody else does
One thumbnail, four different models.
Most tools tie you to whichever provider they resell. Re-Prompter routes each task separately, checks the model can actually do the job before it sends anything, and disables what it cannot do instead of failing after you have waited.
Three steps
Idea in. Thumbnail out. Then argue with it.
The brief stage exists so a human reads what the model decided before an image is paid for.
Say what the video is
One sentence. Add a headline, a subject, a mood if you have them. Or paste a title you already wrote and let Re-Prompter turn it into a thumbnail request.
Read the brief
Subject, scene, composition, camera, lighting, palette, emotion, foreground, headline placement, and what to keep out. Every field is editable. Change one and the artwork changes.
Edit it in English
"Move the presenter right." "Make the background darker." "Replace the office with a trading floor." Paint a mask when the change belongs to one object. Undo, compare, branch.
What is in it
Seven tools, one panel.
Prompt to thumbnail
An eight-stage pipeline: brief, style, persona, image prompt, artwork, typeset headline, save, display. Each stage is separate, so each can use a different model.
Recreate from reference
Reads a thumbnail's skeleton — arrangement, hierarchy, light — and builds something original on it with your subject. An originality slider decides how close.
Natural-language editing
Say the change. Re-Prompter classifies it, routes it to a real edit endpoint or rebuilds the image, and keeps a history you can undo, compare and branch.
Personas and styles
A recurring person who appears across every project, and a reusable visual language with a strength dial and per-attribute switches.
Title generation
Three to fifty titles across twelve strategies, from your topic or from the thumbnail itself. Then shorter, more curious, less clickbait, stronger keyword.
Your model, your key
Five providers built in, plus any OpenAI-compatible endpoint. Test the connection, see what each model can do, and route each task where you want it.
The honest bit
Image models still cannot spell.
Ask one for a headline and you get THUMBNALL more often than anyone admits. So Re-Prompter generates the artwork with no text at all, keeps the space clean, and typesets your headline on canvas — font, weight, stroke, shadow, plate, position, wrapping — where the characters are the ones you typed. AI-rendered text is still there for when the wording does not matter.